This invention relates to the field of digital data transmission, and particularly to a method of compensating for baseline wander in a transformer coupled transmission link.
When transmitting data from a transmitter to a receiver, bandpass limitations can cause a phenomenon called baseline wander, which is the shifting of a central e.g. neutral, voltage or current value about which data signal excursions occur, positively or negatively from the proper neutral or central value. As will be described below, this can cause jitter or data loss.
With reference to FIG. 1, a schematic diagram is shown as a Norton equivalent circuit, which includes a transmitter 1 comprising a current driver, which may, depending on the network standard, be as simple as an on/off switch or as complex as a linear driver or a Digital-to-Analog Converter (DAC). A receiver 3 (shown as an amplifier) may be as simple as an error amplifier or as complex as a full analog adaptive equalizer. In general, both the transmitter and receiver are terminated by resistances that match the characteristic impedance of twisted-pair cabling 5 which is connected between transformers 7 and 9 which couple to the transmitter 1 and receiver 3. The transformers are included both to block DC currents and to minimize common-mode coupling from the transmitter onto the cabling, thus minimizing EMI radiation from the system.
In an AC-coupled data transmission systems such as that of FIG. 1, there often is the possibility of a phenomenon called Baseline Wander, also called BLW. It is caused by the high-pass nature of the AC coupling, and appears as a data-dependent drift of the theoretical zero-crossing point of the data over time. FIG. 2 shows the effects of BLW for Non-Return to Zero (NRZ) data.
The top trace in FIG. 2 is the ideal differential signal that would be expected by the receiver without AC coupling. The second trace shows how this is actually encoded as a pair of differential signals arriving at the receiver. The third trace shows the effects of AC coupling on these differential receiver inputs; as the data pattern density moves from balanced data on the left to unbalanced data (more xe2x88x921""s than +1""s), the signals will start to collapse toward one another. As a result, the overall differential signal""s baseline will wander, as shown in the final trace.
For non-return to zero (NRZ) data, where there is only a single threshold required to determine whether the signal transmitted was a +1 or a xe2x88x921, the effects of BLW are comparatively minor. If the receiver slices the data using a slightly wrong threshold, the only effects appear as edge jitter. For multi-level (Pulse Amplitude Modulated or PAM) codes, however, where there are multiple thresholds required to resolve the transmitted symbol, the effects can be catastrophic and result in data errors. As a result, compensation for BLW becomes important.
BLW may be somewhat controlled by maximizing the open-circuit-inductance (OCL) of the transformers used. However, transformers act as band-pass filters. The low-frequency cutoff is controlled by an induction/resistance L/R time constant, where L is the OCL. A larger L provides a lower frequency cutoff.
The high-frequency cutoff is controlled by the coupling coefficient of the transformer and the same L/R time constant. An ideal transformer has a coupling coefficient K of 1, while real transformers have a coupling coefficient somewhat smaller than, but still close to, 1. The high-frequency corner of the bandpass response has a proportional 1/(1xe2x88x92K2) term in it, so the closer K is to 1, the higher frequency this corner is. It is difficult to manufacture transformers with wide frequency responses (i.e. K close to 1), so when attempting to transmit high data rates, designers are forced to use transformers with the minimum OCL that they can. As a result, designers are competing against two mutually exclusive requirements: high speed requires low OCL""s, while BLW immunity requires larger OCL""s or more expensive transformers. By compensating for BLW somewhere other than in the transformer, the designer is free to use less expensive low-inductance transformers to achieve high speeds.
The simplest way to compensate for BLW is to choose a transmission coding scheme to remove all spectral components from the transmitted data that are below the high-pass limit of the transformer coupling. Among others, transmit line coding techniques known as Return-to-Zero Alternate Mark Inversion (RZ-AMI), Modified Frequency Modulation (MFM, also known as Manchester Coding), and the technique described in U.S. Pat. No. 5,200,979 all do this at the expense of increased bandwidth or required signal to noise ratio (SNR). The MLT3A technique described in U.S. Pat. No. 5,655,078 also removes low-frequency spectral components, but re-uses the increased SNR required by MLT3 and so is not quite as expensive as the other techniques.
Another way to compensate for BLW is to scramble the data with a known Pseudo-Random Bit Sequence (PRBS) before transmitting it, then descramble it at the receiver. This technique is based entirely on statistical arguments based on the length of the PRBS sequence, since there is a small but finite probability that transmitted data will exactly match up to the inverse of the PRBS sequence over a long enough period of time to induce a BLW event.
In the Fiber Distributed Data Interface (FDDI) standard, there is a known valid xe2x80x9ckiller packetxe2x80x9d, that, if transmitted, will induce a BLW event that will break the network in the absence of BLW correction. Other standards, such as 155Mbit/s ATM and 100 BaseT Ethernet may also suffer from BLW xe2x80x9ckiller packetsxe2x80x9d, but this is a matter of some debate.
Yet another way to compensate for BLW is to include BLW correction circuitry in the receiver. This requires that there be circuitry built to detect the BLW event and add an offset to compensate. Depending on the complexity of the rest of the equalizer, this may be easy or difficult. As an example of the trade-offs required, FIGS. 3A, 3B and 3C show three possible BLW correction loop configurations with an adaptive equalizer 11 receiver. The first, in FIG. 3A, shows a BLW correction loop 13 applied before equalization. This is, in theory, the simplest to conceive; however the BLW correction algorithm in the loop needs to extract information from the signal before the equalizer, so any correction algorithm that assumes a known decision out of the equalizer is impossible.
The second configuration, as shown in FIG. 3B, wraps the BLW correction loop 13 around the adaptive equalizer 11, in effect embedding the adaptive equalization loop as an inner loop of the BLW loop. It has the advantage over the first configuration in that now the BLW correction algorithm may use the information coming out of the equalizer as its input. However embedding the adaptive algorithm loop as part of the BLW loop makes stability analysis of the BLW loop challenging.
The third configuration shown in FIG. 3C puts the BLW correction 13 after the adaptive equalizer 11. This has none of the difficulties of the first or second configuration, but requires that the entire equalizer have an increased dynamic range (as much as twice if the transformer voltages completely decay) to accommodate the extremes of a BLW event. As supply voltages are reduced with smaller-geometry microelectronic processes, this linearity becomes harder and harder to accomplish.
Yet another technique i$ to include BLW correction circuitry into the transmitter. Because in this technique the BLW correction circuitry is a single loop, not wrapped around an adaptive equalizer, it is much simpler to analyze and design than the receiver BLW correction circuitry of FIG. 3B. At the same time, because the transmitter comes before the cable 5, the BLW correction algorithm already has information about what signal is being transmitted and doesn""t have the difficulties of the design of FIG. 3A. Finally, since the transmitter minimizes the BLW of its output, the receiver needs only enough dynamic range to accommodate the signal. It does not need the excess dynamic range required by the circuit of FIG. 3C.
Reference is made to U.S. Pat. No. 5,655,078 by Paul M. Anderson et al, issued Aug. 5, 1997. This patent describes MLT3A, a variant of MLT3, in which a memory is kept of the transmitter""s history and an attempt is made to ensure a long-term balance of +1""s and xe2x88x921""s on the cable, thus minimizing BLW. It is, however, an open-loop solution and is not easily generalized to other line coding techniques.
Reference is also made to U.S. Pat. No. 5,465,272 by Robert W. Smith, which describes a system readily generalized to a wide number of line coding techniques. It embodies a single loop in the transmitter, using a ternary winding on the transmit transformer to sense the signal being transmitted on the line. It then compares the sensed signal to the idealized transmitted signal, and then applies a correction signal to suppress any BLW present in the output. A practical implementation of the concept shown in FIG. 2A of the patent is shown in FIG. 4 herein.
The main driver 15 and correction driver 17 are controllable current sources whose outputs sum to drive an off-chip resistor 19, the primary winding of transformer 21, and via the secondary winding, the twisted-pair cable 5. A reference driver 23 (matched to driver 15) and an on-chip resistor 25 (matched to resistor 19, the transformer 21, the cable 5 and the far-end receiver termination) generate an ideal on-chip output voltage without baseline wander. A ternary winding of the transformer 21 and a second off-chip resistor 27 connected across the ternary winding sense the voltage on the line and any BLW on it. Error amplifier 29 compares the reference output to the sensed voltage on the line, then, through low pass filter 31, drives correction driver 17 to correct for the BLW at the outputs.
The cost associated with adding the transmit baseline wander correction circuitry of FIG. 4 to the circuitry of FIG. 1 has several components. A simple transmitter without BLW correction only consists of driver 15, resistor 19, and a two-winding transformer 7. Adding transmit BLW correction requires elements 17, 29, 31, 23, 25 and 27 and a three-winding transformer 21. Elements 17, 29, 31 and 23 together with element 25 are all on-chip and as a result cost power and on-chip layout area, but can have a low dollar cost. Resistor 27 has a low dollar cost, but doubles the printed circuit board area taken up with transmitter termination resistors and increases the power that the chip needs to drive into off-chip circuitry. The three-winding transformer 21 is larger and occupies more printed circuit board area than a two-winding transformer. Because of a lack of economies of scale, the three-winding transformer would also be significantly more expensive than the much more common and easier to obtain two-winding alternative.
In addition, the sensing feedback path from the transformer back to the chip is an additional high-speed path that would have to be made with controlled-impedance printed circuit board traces, doubling the board area taken up with transmitter signal routing. Finally, the sensing feedback path requires two additional chip pins, which on a multi-channel device can be a significant additional cost and may force the chip into a larger and more expensive package.
The on-chip costs (elements 17, 29, 31, 23 and 25) are, for the most part, easier to justify than the off-chip costs. Because lumped series resistance of the twisted-pair cable limits the low-frequency correction possible by this technique, driver 17 can be made significantly smaller than driver 15, both in power and area. Driver 23 and resistor 25 can be scaled copies of driver 15 and resistor 19, reducing their power and area. Because of the limited bandwidth required, error amplifier 29 can be made small and low-power, while the corner frequency of filter 31 does not have to be particularly accurate. There is, however, an additional major problem. Because lumped series resistance of the cable (or mismatches in the transmit or receive terminations) will introduce a gain error in the feedback path, either driver 23 or resistor 25 needs to be variable with an adaptation loop to compensate for this gain error. If this is not done, this BLW correction loop will, in effect, introduce BLW where there was none before. This adaptation loop would need to monitor the incoming data stream, looking for a long run of balanced data which will have minimal BLW when transmitted. It then must disable the BLW correction loop and trim driver 23 or resistor 25 as necessary to match the amplitudes transmitted to those received, and then re-engage the BLW correction loop, remembering the trimmed value. This loop is fairly complex and will increase the on-chip complexity considerably. Nonetheless, the on-chip costs are small compared to the off-chip costs.
The off-chip costs (involving resistor 27, a more complex transformer 21, additional area taken up with routing, and the additional two pins) are significant and as a result would require that this style of transmit BLW correction have significant benefits in order to justify it.
If a transmit BLW correction loop could be designed which has substantially the same off-chip cost as the simple transmitter, it would be much more likely to be used. In addition, if the adaptive loop to compensate for lumped series resistance of the cable (or termination mismatches) can be simplified or removed, the correction technique would again be much more likely to be used.
The present invention provides a method and apparatus for providing baseline wander compensation in a transmitter which has substantially the same off-chip cost as the simple transmitter, and in addition, does not require the adaptive loop to compensate for lumped series resistance of the cable, or termination mismatches, required in the prior art circuit of FIG. 4. Only a two winding transformer is required. As such it provides a significant improvement thereof.
In accordance with an embodiment of the invention, a method of compensating for baseline data signal drift of an ongoing signal comprises sampling the signal and holding the sample, comparing the sampled signal with the ongoing signal at a time later than the sampling time and deriving a difference signal, and correcting the ongoing signal in a direction so as to substantially eliminate the difference signal.
Sampling is effected from the output signal, and the transformer is a two-winding transformer which need not have a ternary winding. Sampling can be done across the primary winding, or the secondary winding with appropriate coupling circuitry.
In accordance with another embodiment, a circuit for correcting baseline wander in a transmitter which is comprised of a main signal current driver driving a first winding of a two winding transformer which first winding is in parallel with a termination impedance matching load, comprises:
(a) a compensation current driver coupled in parallel with the main signal current driver for adding its current to current of the main signal current driver, and having a correction signal control input,
(b) a sample-and-hold circuit connected across the first winding,
(c) an error amplifier having a first pair of inputs connected across the first winding and a second pair of inputs connected to the output of the sample-and-hold circuit, for providing an error signal resulting from a difference of signals appearing at its respective pairs of inputs, and
(d) a circuit for applying the error signal to the correction signal control input of the compensation current driver,
whereby any difference of a present signal carried by the first winding of the transformer, with a past signal held by the sample-and-hold circuit, results in an error signal and generation of a correcting current by the compensation current driver so as to substantially eliminate the difference.
In accordance with another embodiment of the invention, a method of compensating for signal drift comprises applying an output signal and a representation of an input signal to respective inputs of an error amplifier, applying an error signal resulting from an output of the comparator to a control input of a signal compensation driver, generating the output signal from an input driver driven by the input signal and from the signal compensation driver, and applying the output signal to a single winding of a two-winding transformer.
In accordance with another embodiment of the invention, a circuit is provided for correcting baseline wander in a transmitter which is comprised of a main signal current driver driving a first winding of a transformer,
(a) the transformer being restricted to having two windings, and further comprising:
(b) a termination impedance-matching load in parallel with the first winding of the two winding transformer
(c) a compensation current driver coupled in parallel with the main signal current driver, having a correction signal control input, for adding its current to current of the main signal current driver,
(d) a reference current driver having a control input coupled in parallel with a control input of the main signal current driver for receiving an input signal,
(e) a reference resistor connected across the output of the reference current driver,
(f) an error amplifier having a first pair of inputs connected across the first winding and a second pair of inputs connected across the output of the reference current driver, and
(g) a circuit for applying the error signal to the correction signal control input of the compensation current driver.